Next
Page >>
The Development of English Word Order:
A Minimalist Approach
Joseph Galasso
California State University,
Northridge
2001
Introduction
This
paper is devoted to examining the variety of word ordering found
in the English data. In §1.1 I present two rather different accounts
of variable Word Orders found at the proposed prefunctional stage
of language acquisition: (i) Tsimpli (1992), and (ii)
a weakened version of Kayne (1995). The main thrust behind the
two proposals leads me to conclude that at the very earliest stage
of language development, there seems to be some evidence to suggest
that some aspects of language principles (typically those associated
with LAD) might be subject to maturation (§1.2). In §1.3 I examine
the data.
(Preview)
Following
Kayne's LCA, I come to the conclusion that Word Order is indeed
a universal hierarchical property of a Spec>Head>Comp relation.
However, this, in itself, is not enough to account for the wide
array of word order errors found in the early data. Adopting Kayne's
Head Medial Principle, I consequently devise a two-pronged
model for dealing with the variable word order patterns found
at this initial Two-Word phase of development--referred to here
as SAS=Single Argument Strings (viz., the phase at which children's
speech chiefly consists of one argument; cf. Bowerman 1990). The
core assumptions underlying this model are the following. First,
that all Heads must be base-generated in medial position. Second,
that there exists a Complement Parameter which endeavors
to set the Object--here, interpreted as being initially hosted
within a non-agreeing Spec position (cf. Larson)--either leftward
or rightward of the Head. As a result, this parameter leads to
the eventual correct positioning of the agreeing Specifier
that must then enter into an agreement relation with V within
a given DAS structure. I subsequently conclude that it is the
onset of the DAS (=Double Argument String) phase which triggers
not only Kayne's weakened version of LCA, but, more theoretically,
establishes the (Larsonian) VP-shell which provides important
A'-bar positions to movement. In one sense, the core of the SAS-phase
could be considered as the lexical-thematic stage par excellence.
1.1 Word Order
Most children pass through what seems to be a two-word phase.
What makes this phase interesting is the observation that rarely
more than a maximum of two words (Head and Argument) ever appear
to project within any given utterance: viz, both transitive and
di-transitive verbs (e.g., kiss, give, put) appear at surface
syntax without one (or more) of their required internal arguments [1]. A number of researchers (notably
L. Bloom 1970, Braine 1976, Bowerman 1973, 1990, and Hickey 1992)
have claimed that a correlation seems to hold between the onset
of the Verb's second argument and the advent of fixed target word
order. With respect to these claims, the following sections proceed
to examine Word Order at the two and three word
phases of development.
Recall that in the syntactic framework being adopted here the
Subject of the target syntax is assumed to raise into Spec of IP
(a functional projection) where it proceeds to check both T(ense)
and Agr(eement) features along with its Nom(inative) Case feature.
This checking is done via a local Spec-Head relation in IP: (In
English) the features of the Head-Verb raise covertly from
the Spec-VP (at LF) while the Spec-Subject raises overtly
(at PF). The Object(s), on the other hand, is assumed to remain
in-situ within the Complement of V' where it receives its
inherent Q-role
assignment along with its Objective Case via the lexical Verb. In
this way, movement operations can be defined as being associated
with Functional Categorial/INFL projections: within the Structure-Building
model being proposed here, such operations should therefore
be excluded at the Pre-functional/Non-INFL-stage. It follows then
that at the prefunctional stage any variation of word ordering (found
in corpora of child language acquisition) should represent base-generated
orderings.
The following two analyses offer formal accounts of observed word
order variation found at the lexical stage, and accordingly, serve
as theoretical foundations for how I later account for my own Data
presented below.
1.1.1 Word order variation at the VP-Stage: Tsimpli
(Proposal I) Tsimpli (1992) assumes an early Prefunctional/Non-IP
structure as in (1) below (irrelevant details omitted) characterized
by free word order at the Lexical-Thematic VP stage. Specifically,
SV(O) and V(O)S are seen as possible patterns, while other orderings
(e.g., VSO and OSV) are, in principle, ruled out as being base-generated
(for reasons having to do with UG requirements on predication which
stipulate (inter alia) that the Subject cannot intervene between
the Head and its Complement). [2] The data below present only the
variant VS, OV ordering (taking correct SV, VO for granted):
(1')
English VS (Bowerman 1990)
a. ride Christy
[-Finite]
b. write Sissy
c. drink mama
d. open Christy
e. fall mommy
f. cut boy
g. going daddy |
French VS (Pierce 1989)
h. lit mama
[+Finite]
(reads mommy)
i. travaille papa
(works daddy)
j. ranger moi
[-Finite]
(arrange me)
k. tomber pap
(fall daddy)
l. dormir bébé
(sleep baby) |
The structure in (1) (with token declarative examples of VS in
(1-prime) above) permits the subject to be base-generated--receiving
its external Q-role--either
to the left or right of the V'. This assumption of a free word order
at the VP-stage is based on the notion that only functional categories
are responsible for Directionality setting (via FPH). In the above
sense, a specific directionality for Q-marking
is not universal and, therefore, must be set via parameterization.
[3] Since only (non-parameterized) substantive
Lexical categories appear at the VP-stage, it then follows that
there should be, in theory, no Directionality restrictions. Moreover,
in assuming that the VP-stage can host only thematic argument positions
(cf. Radford 1990), there are then, in principle, no non-thematic
argument positions to serve as potential landing sites for a moved
argument (cf. Q-Criterion).
Therefore, the structure in (1) is consistent with the notion that
Verb and Subject raising would be inoperative at the Lexical Thematic
VP-stage. Again, this amounts to saying that any word order pattern
observed at the VP-stage would, by definition, involve a base-generated
pattern. However, Tsimpli adds an appropriate caveat here, noting
that such a 'permissive' word order stage doesn't necessarily result
in all possible word combinations being equally attested at the
VP-stage.
Tsimpli goes on to look at a range of language acquisition data--all
of which seem to permit free word orders at the VP-stage--in support
of the structure given in (1). An interesting note here is that
in her examination of French (Pierce 1989), she seems to find no
correlation between the usage of correct ordering and the emergence
of the Verb's Finiteness feature (a traditional functional feature).
In other words, both SV and VS word orders are attested within finite
structures. However, the notion of a Finiteness feature projecting
an extended Functional Phrase (=FP) above VP at this early stage--deriving
a VS pattern via verb raising--is ruled out on the basis that the
frequent occurrence of postverbal Subjects would then be unaccounted
for. (E.g., why would only Verbs raise leaving Subjects in-situ?)
It had been assumed earlier that Word Order was indeed determined
by such a functional category--the question to ask here is then
the following: why doesn't the functional category associated with
the Finiteness feature determine word order here? In answering,
Tsimpli claims (Tsimpli et al. 1995) that the early emergence of
strong morphological Agreement features does not necessarily correspond
to a full projection of X-bar structure. She remarks:
Early data from Greek and Spanish show that agreement morphology
is available from the earliest appearance of verbal forms. Similarly,
aspectual distinctions in Greek, Spanish, Irish, German and English
appear to be marked in a way similar to the corresponding adult
grammars. From this, however, it does not follow that the corresponding
syntactic categories have matured in the sense that they are able
to project the relevant X-bar structure (op.cit: 25). [4]
The above observation is consistent with a Structure-building
account of language acquisition in the following manner: viz.,
if the acquisition of functional categories is based on an innate,
maturational program, then morphological triggering alone might
not be a sufficient account of how the child comes to acquire the
relevant extended phrase structures. Tsimpli also takes into account
the problem of how to define morphologically rich vs. impoverished
morphology. [5]
In this next subsection, we look to how Tsimpli examines the syntax
of Object-Verb strings within the same configuration that was used
to discuss the above Subject-Verb Strings. Hence, following Tsimpli,
I draw the same tree twice for two different constructs (noting
that now NP (=Object) is positioned V' internally):
(2')
English OV
(Bowerman)
a. horsie ride
b. daddy see
c. dolly hold
d. bottle find
e. salt pour
f. milk drink
g. going daddy |
French/Spanish OV
h. chapeau chercher
(Clark 1985)
(hat look-for)
i. let li
(letter read)
j. agua beber
(Pina 1984)
(water drink)
k. sillon abre
(chair open) |
The structure presented in (2) similarly assumes the Object to
be base-generated in one of two positions where it can receive its
Q-role
assignment via the Verb. Such a structure permits both OV and VO
patterns. Tsimpli, citing a variety of literature on the instability
of early Head-Complement ordering for target SVO languages, reports
that VO/OV patterns are not only widely attested at the two word
stage, but that a correlation seems to exist between such instability
and the omission of Subjects. In other words, (citing Bowerman 1973,
Braine 1976), when Subjects and Objects simultaneously enter into
a projection, the SVO pattern often predominates (while acknowledging
that there may be a transitional period--give or take a few weeks
between the SAS and DAS phases--where variable orderings may persist).
Tsimpli goes on to cite Bowerman's (1990) observation that a correlation
is found between variable orderings and the Single Argument String
(=SAS) phase (e.g., OV/VO, SV/VS) (NB. Bowerman gives no mention
of a possible two-word (DAS) phase yielding SO/OS structures.
This might mean that the verb plays an essential role in word order
(though see note 1)).
Similarly, Hickey (1992:11) shows that frequent incorrect *SV
orders for early Irish fall away with the onset of DASs (The Eoin
Data reports n.=86 VS compared to 14 *SV frequency counts
within Finite SAS constructs as compared to a n.=0 *SVO count.)
Such findings could suggest that correct word order seems to emerge
just after the onset of the Double Argument String (DAS) phase.
Tsimpli suggests that the correlation might have something to do
with the idea that universal semantic relations, expressed by fully-fledged
Thematic Roles at the DAS phase, determine positional patterns.
This would therefore suggest that the DAS phase represents what
could be considered as the parameterized stage. (NB. A word of caution
here is in order--up until now, the functional parameterized stage
has never been clearly associated with the DAS phase.) Regarding
the hypothesized unparameterized SAS phase, Tsimpli's claim is further
enhanced by the unlikelihood that directionality or linearisation
of Q-marking
ever acts in a universal manner. Thus, in assuming that the direction
of theta-role assignment varies from language to language, there
is no reason to assume a universal directionality of argument structure
for e.g., Put to be [Put: agent <theme, location>]
in that order. Likewise, I see no inherent reason why e.g., an external
argument should require to be projected to the left of the Verb.
In this sense, the directionality of Theta role assignment indeed
does come under parameterization (at the DAS phase) (but see Radford
(1990:45) for a different view which proposes an inherently fixed
word order at the lexical-unparameterized stage). Whatever the case,
the correlation stated above suggests that it is the simultaneous
projection of both SPEC and COMP that bring about Q-Directionality.
In assuming the two structures presented in (1) and (2), un-parameterized
principles of UG permit the following array of base-generated
word orders: SV, SVO; VO, VOS; OV, SOV, OVS. However, as deduced
from (1) & (2), VSO and OSV orders--where the Subject is seen
as intervening between the Complement and the Head--are unavailable
as base-generations. As we shall see in the following section (§1.2),
the number of possible UG base-generated patterns becomes further
reduced--an outcome much sought after for Learnability's sake.
1.1.2 Word order variation at the SAS-Phase: Kayne
Proposal II Before turning to my Data on Word Order variation
found among Single Argument Strings (SASs), let us take a moment
to consider briefly how such variable orderings might be accounted
for within more recent interpretations of Chomsky's Minimalist
Program (Chomsky 1995). In Kayne (1994), attempts are made
to strengthen X-bar Theory in such a way that a Universal base-generated
ordering of Specifier-Head-Complement (SVO) is maintained--all
other word orderings being derived via movement. (NB. Chomsky endorses
Kayne's modification of X-bar (1995:334-340) albeit with some refinements
having to do with Bare-Phrase Structure. See §1.4). It seems
that such a tightly restricted theory would undoubtedly encounter
empirical problems regarding the protracted nature of language acquisition.
More specifically, how might Kayne's model handle apparent word
order errors found at the very earliest SAS Non-INFL Stage-1;
a stage where (Argument/Head)-movement operations and, arguably,
A-bar positions are not yet available?
In Kayne's model, the Complement position plays a central role
in determining order (via hierarchy): i.e., within the Head-Comp
configuration, it is solely due to the Complement's maximal projection
that strict linearisation--imposed by the LCA--is maintained. The
bases of Kayne's model bare some resemblance to traditional distinctions
between the Specifier-Head relation--being an external
argument relation--and the Head-Complement relation--being an internal
relation. Such distinctions, stemming from traditional Case and
Q-Theory,
basically state that a natural asymmetry holds between the Subject
(which gets Q-role
assignment from within the VP, then, via movement into Spec of IP,
checks its Case in a Structural relation with INFL outside VP's
maximal projection), and the Object (which inherently receives
Case and Q-role
assignment via a sisterhood relation with V directly, and inside
the VP maximal projection). This traditional Spec-Comp asymmetry
is even more pronounced given Kayne's recent interpretation that
redefines all proto-type Specifier positions as Adjuncts. Presenting
straightaway Kayne's model, let's consider (3), (3-prime) and (4)
below (1994:7,16):
(3)
In (3), it is precisely the insertion of a maximal-projecting
N node of the Complement that breaks the symmetry of L [M,P;m,p]
found in (3') resulting in an antisymmetric hierarchy. (Without
going into further detail here, one property of the linear order
found in (3) is based upon The Linear Correspondence Axiom
(LCA). [7]
In short, the axiom states that an asymmetric c-command relation
(=ACC) imposes a linear and hierarchical ordering of terminal elements
(from left-to-right): e.g., terminals K, L, N
as seen in (3). The central workings of ACC lie within the Head-Complement
relation, and for that reason alone an additional (VP-internal)
maximal projecting node N of the Complement is required--bringing
about an asymmetric left-to-right hierarchical structure. In other
words, the external Specifier seems to play little role in determining
the subsequent asymmetry in (3). However, the Specifier structure
in (3) does pose linearisation problems for Kayne (e.g., regarding
multiple Specifiers, etc.) and, as a consequence, gets refined (in
the way shown in (4)). In short, the SPEC problem is resolved by
restricting c-command to categories only, thus forbidding segments
such as Spec-Adjunct from entering into c-command relations. Hence
in (4), the Specifier (P2) becomes redefined as an (segmental)
Adjunct position (from P1),
thus saving LCA. (An additional stipulation is made which states
that Adjuncts must exclusively adjoin leftward, thus eliminating
potential problems regarding Spec-final structures.)
Some crucial questions to pose here are the following. First,
how might we apply such a seemingly stringent X-bar mechanism for
dealing with very early Single Argument Strings (SAS) that appear
to contradict claims of universality of innate ordering?
[8] Second, if we assume that the Lexical-Thematic
stage is without Functional projections and movement operations
(cf. Radford 1990), how might we save Kayne's model? Well, suppose
that the child initially starts off with just a proto-type Thematic
Argument+Head configuration as place-holders within K (=XP) (as
in (3)) for the Single Argument String, yielding typical SV/OV ordering
(order irrelevant here). In addition, assume that the N terminal
node that is traditionally seen to host the Comp is not yet projected.
In this sense, a potential Specifier/Subject is not yet considered
as an Adjunction position, but rather is seen as an external argument
position of the verb's first argument. We might wish to re-think
this anomalous proto-argument slot in (5) below as either hosting
a Specifier or Complement (Comp=Object in superficial Subject position)--replacing
(for the time being) traditional distinctions laid out between external
Specifier-Subject and internal Complement-Object positions, with
the notion that it is a Unique Specifier slot of the Verb
which projects from out of this proto-argument placeholder. (Note
that this account may also share some similarity to current trends
in Minimalism that claim that Objects indeed enter into the (second)
Specifier position of a Larsonian VP-shell--picking up
appropriate theta-roles along the way en route to Spec-AgrO. ).
[9] Consider the reduced trees in (5) (cf. 3)
where the Spec/Comp is now "reorganized" and fused as
the only potential external argument: [10]
(5)
Not only would such a configuration yield SV and OV,
but likewise (mirrored) VS and VO orderings--a much needed configuration
for dealing with the present data. (It is interesting to note that
such word order variability is likewise restricted to the Head+Proto-Spec
Argument relation in Tsimpli's model above--i.e., arguments can
either precede or follow the Verb).
In (5), two crucial aspects of Kayne's asymmetry are broken down
for the following reasons. First, due to the absence of a maximal-projecting
N node Complement position (c-commanded by L), the sole potential
argument slot does not need to be defined in terms of Adjunction
for reasons alluded to in (4). This enables the child to treat both
Subjects and Objects now as one sole Lexical proto-argument which
is, in turn, Q-marked
by L: recalling the intuitive assumption that, at the lexical-thematic
stage, all Specifier-Head-Complement positions are thematic (and
furthermore, share strict sisterhood relations with V/V'). In example
(5), the Spec/Comp may assume the same syntactic role--where only
a fundamental thematic sisterhood relation holds. The spirit of
this proposal closely follows similar lines previously taken in
Radford (1990), who likewise assumes that in Lexical-Thematic OV
structures, the preverbal Subjectless slot of the Specifier could,
in fact, be filled-in by a preverbal Object. Radford later goes
on to clarify that whenever a subject and object simultaneously
enter into a projection, the object always positions postverbally
(with the preverbal subject) yielding correct SVO. Consider the
token examples in (5') below (Radford 1990:232 citing Bowerman 1973):
(5')
Preverbal Objects Base-Generated at SAS-Stage
a. Doggie sew (=sew doggie)
b. Kimmy kick (=kick Kimmy)
c. Kendall pick-up (=pick-up Kendall)
d. Doggie look-at (=look at doggie)
|
A number of problems however arise from such a Proto-XP scheme.
First, how would we analyze potential pro-subject elements within
OV/VO structures? Since the sole argument-slot is already occupied,
there seems to be no place-holder for a pro element to enter into.
One way around this dilemma might be to stipulate (extending Rizzi's
analysis) that at this thematic SAS phase, all missing arguments/predicates
(such as pro) are potentially lexically saturated (cf.
Rizzi 1986:508) and thus implicit--i.e., their Q-role
assignment may not require a (lexically overt) syntactic projection.
(This would predict that all predicates, among SASs, could allow
null arguments.) Their reference may be determined, and subsequently
handled, purely on pragmatic grounds that do not affect ordering
constraints. In other words, any argument at the SAS phase could,
in principle, occupy the proto argument-slot without syntactically
impeding upon a given pro. A second problem however has
to do with more fundamental notions of Q-role
assignment. Traditional assumptions pin the thematic assignment
e.g., Agent/Patient to the Subject, via an externalizing
mapping role: i.e., as mentioned above, Subjects are distinct from
Objects in that Subjects reside exclusively outside the positions
of the V-bar (Objects remaining within the direct sister relation
of V-Head). In the proto XP tree (as presented here in (5)), there
is no distinction between external and internal arguments: i.e.,
both Subject and Object are analogously projected. [11]
This presents classic Discontinuity as well as Learnability
problems for the child. More specifically, (i) how might
the child determine the correct theta role assignment under such
sparse conditions; and (ii) how then might the child "Delearn"
this strategy once the proper fully fledged binary branching structure
is achieved? One suggestion that has been proposed in the literature
has to do with the notion that some visibility mechanism
must first be achieved by the child before any assignment of Q-role
can be determined. Radford (1990:245) suggests that an additional
externalization mechanism is required (i.e., must await parameterization)
in order to determine which theta-marked argument of a predicate
will be projected in the syntax as an external subject (in Specifier
position). [12]Accordingly,
we take the standard view (e.g., Williams 1981) that the Subject/Specifier
is unique in that it exclusively selects the external argument position--however,
with the exception (being proposed here) that this uniqueness
property only holds with respect to Double Argument String environments
(i.e., Subjects acquire this inherent externalization mapping only
via their opposition to a given Object). This amounts to saying
that a hierarchy of Q-role
visibility exists: (i) the first argument of a verb is
given a proto external argument position, (ii) once a second
argument of the verb is syntactically projected (catapulting the
full-fledged XP), both externalization/internalization mechanisms
become operational forcing the Subject/Specifier now to solely occupy
the external position--leaving the Object to insert under the Complement.
Such a hierarchy seems to be consistent within Minimalist assumptions
where the Head-Complement relation is not only "more local"
but also more fundamental--associated with thematic relations, and
where the Specifier-Head relation is considered to be an elsewhere
category (Chomsky 1995:172). In one sense, this seems to match
what we are saying. First, the proto-argument slot here is rather
syntactically undeveloped and does not seem to adhere to the full
range of syntactic relations as expressed above. Furthermore, since
this structure involves the thematic stage anyway, theta-roles will
be expressed in any event (irrespective of the more local
relation of Head-Comp as defined above). Second, the emergence of
Object (second argument) as distinct only from the Complement position
now establishes the kind of relations mentioned above: namely, where
the Head-Comp is seen as being fundamental (the cornerstone of thematic
relations) and where the Spec can now be defined as the elsewhere
category. Hence, the basic intuitions along these lines are as follows.
(i) The notion that the internal Complement acts in a fundamental
way--fulfilling its epistemological priority role and taking
precedence over an external position--holds only for DASs; where
the full-fledged dual-argument projections of XP are developed.
(ii) It is only when a verb's second argument overtly enters
into a syntactic projection that the Subject/Specifier becomes distinguished
as the legitimate external argument. This view is consistent with
Radford (1984) who claimed for a Uniform Bar Expansion Hypothesis--whereby
the first stage in the production of any category was to project
only a Head X; then to add a single argument projecting X into X-bar;
and finally adding a further argument projecting X-bar into XP.
What I am saying here is similar, though arrived at via a different
route.
In sum, the basic claim presented thus far is that very early
child two-word SAS utterances are to be analyzed in terms of Predicate-Argument
XP constructs only--where the Predicate is seen as a Proto-Head,
and the Argument as either a Proto-Specifier or Proto-Complement
(both occupying the external argument position). Hence, in the earliest
SAS phase, there is continued support toward maintaining those primitives
of X-bar Theory (albeit a reduced XP in the sense that the internal
argument doesn't project), which pertain to distinctions of Headness
and External argument (Non-headness). However, no support (at the
SAS phase) is gained towards the notion of an innately given fully
expressed XP hierarchy. [13] The refinement of Kayne's model
would then predict that at the very earliest two-word SAS phase,
children make salient a variety of base-generated configurations.
Such a model (based on LCA) would go on to predict that once Double
Argument Strings (DAS) emerge, a strict SVO pattern would develop.
This is generally borne out on both Theoretical and Empirical grounds.
Theoretically speaking, once ordinal Specifiers and Complements
simultaneously enter into a projection, two crucial things happen.
First, in order to save Kayne's hierarchical ordering, the Complement
is stipulated to have an asymmetrical maximal projecting terminal
node (N) breaking the symmetry of L in (1). Second, since
the Complement projects, the Specifier must now be redefined as
an Adjunct both in order to save ACC and to prevent unwanted Spec-final
outcomes. Both conditions are now in place yielding Kayne's Universal
SVO ordering (OVS are theoretically ruled out by Kayne on rather
abstract reasonings--however, see (§1.2) on OVS as a potential
Universal ordering).
Empirically, this seems to be what one finds for English. Citing
a number of data, a correlation seems to exist between fixed word
orderings and DAS constructions (cf. L. Bloom 1970, Braine 1976,
Bowerman 1973, 1990, Deuchar 1993, among others). Radford (op.cit:232)
(citing Bowerman 1973) demonstrates that though we indeed find alternations
between SV/VS and VO/OV orderings, we rarely find instances of SOV
for English at the thematic stage (being an ordering derived via
movement). [14] (Tsimpli (1992), on the other
hand, does find such orderings though in a limited way compared
to the vast variability found among SASs (see note 6)). Once a base-generated
Subject/Object syntactically projects, the Subject/Object is always
forced to remain in its respective Spec/Comp position. (This is
also evident from early Wh-Question CP>VP constructions (without
Do-insertion or Aux-inversion), where the Wh-element is assumed
to raise (via O-movement) from out of the Complement of VP into
Spec-CP. Word Order errors within the VP therefore would be predicted
not to occur within Wh-constructions.
In sum, the following abstracts of the two phases correspond to
earlier cited data (cf. (1) & (2)--Bowerman, Bloom, Braine,
Tsimpli, ibid) regarding variable ordering at the (Non-IP) SAS phase
as opposed to the DAS phase:
Abstract-Summary of SAS
vs. DAS (Bowerman 1990) |
SAS Phase
1. 60% of SAS strings with proto-typical agent-patient verbs
were mis-ordered.
2. Both OV and VS orderings of proto-typical agent-patient verbs
occur.
3. Postverbal subjects appear with transitive, intransitive
and unaccusative verbs.
4. It is concluded that both pre/post-verbal
subjects are base-generated--not derived. |
DAS Phase
5. There is a 'transitional' period lasting a few weeks where
DAS permit variable orderings. After this period, (at approx.
2 years of age), the order of DASs become fixed to target SVO.
6. Among DAS utterances, 22 tokens of correctly ordered VO strings
were found--not one single token of incorrect OV order. |
Summary: word order is unstable for single argument strings--yielding
SV/VS, VO/OV. |
Summary: word order becomes fixed (after a brief transitional
period) to target SVO with the onset of double argument strings. |
1.2 SVO and OVS
Much of Kayne's model of a Universal Hierarchy hinges on theory
internal stipulations. For instance, Kayne (op.cit:35-38) recognizes
that a potential OVS ordering could be consistent with LCA: e.g.,
Kayne here is further required to explain why <x, y> is interpreted
as 'x precedes y' rather than as 'x follows y'. Kayne's methods
of argumentation here seem rather abstract and circular, relying
on an assumption that universal temporal orderings flow from left-to-right.
For the sake of argument, let's suppose that Kayne fails to convince
us of this one point and that SVO and/or OVS (its mirror
image) constitute potential UG orderings. [15] Samuel Epstein
(cited in Chomsky 1994a:fn.32, p.437) is equally unconvinced and
notes that Kayne's abstract formulation allows very free temporal
orderings even if LCA is satisfied. Thus, if a class of phrase markers
satisfies LCA, so will any interchange of sisters. Thus, we
are unavoidably faced with any arrangement of orderings within the
Head-Complement relation (e.g., read-books or books-read freely).
Kayne himself remarks that though both Head-Complement and Complement-Head
orders are widely attested among the world's languages (as it is
with language acquisition), once the Specifier projects, a much
more visible asymmetric SVO ordering occurs (op.cit: 35). Notwithstanding
such uncertainties involved, one interesting note to keep in mind
here is that a persistent necessity holds throughout for the Head
to position itself medially--i.e., Specifiers and Complements
must project from opposite sides of the Head, all other possible
configurations being derived via movement. [16] The above weakening
of Kayne's UG hierarchy has important implications for Child Language
Acquisition cross-linguistically. For instance, in the event that
both SVO and OVS orderings were potentially licensed by UG, this
might allow us access to a much more straightforward account of
base orders for verb-final languages such as e.g., Japanese: languages
that would otherwise (under the SVO account) be forced into making
a number of unparsimonious movement operations. Moreover, the SVO/OVS
UG ordering has much in common with what is being proposed here
concerning the SAS phase. Namely, if we assume that the Specifier
slot is an anomaly, representing the sole argument position at the
SAS phase--the Subject/Object being reorganized within the Proto-Argument
placeholder--then, only a dual-variable of word orders
ever project: (i) between argument-final VS/VO structures
(yielding potential target OVS, SVO grammars respectively) and (ii)
between argument-initial SV/OV structures (yielding the same potential
target SVO, OVS grammars respectively).
Furthermore, we might wish to adopt Kayne's initial observation
that the Head be required to position medially (a fact which seemingly
falls directly out of LCA) and perhaps claim it as an even more
fundamental principle of syntax: [17]
(6) Head-Medial
Principle = <x, y / y, x>.
(Kayne 1994:35)
That is, in more concrete terms, the Head <x> can project
(universally) either in an initial or final surface position in
relation to the Complement <y> prior to the onset of the Specifier
<z>. Such a principle could be reduced to saying that once
the fully-fledged (external) Agreeing-Specifier projects,
it maintains the option of either projecting to the left or to the
right of the head <x> (where it then enters into a proper
agreement relation) depending where in the input the Comp(lement)
positions itself: thus yielding either SVO (if Comp projects rightward)
or OVS (if Comp projects leftward). Recall in (6) that Kayne stipulates
that <x> must remain within a medial base-generated position.
In the light of our refinement of Kayne's mechanism for universal
word order, how then might we eventually set the Specifier position?
Well, although Kayne wishes somewhat to distance himself on the
matter of Word-Order/Directionality via Parameterization, we may,
however, opt to keep traditional notions based on Principles
and Parameters. We might suggest that the Directionality
Parameter depends upon the placement of the Complement.
Hence, Directionality is set via the following parameterization:
[Comp-initial/Comp-final], and thus Spec positions itself accordingly--keeping
to the Head- Medial Principle as stated above. (We can now dispense
with all notions of Head and Specifier Directionality Parameterizations
since, in the model proposed here, their directionalities naturally
fall out from the placement of Comp.)
To make matters more concrete, e.g., languages that show a strong
tendency toward SOV--where Complements are observed in the child's
input to appear to the left of the Head--the base-generated OVS
order might be selected. This would result in forcing the Subject
to raise to the outermost Spec position. Such an approach for Japanese
has recently been suggested (see Tonoike 1993, 1995). [18] In languages, such as Irish
and Welsh, with a strong tendency toward VSO (where Complements
are generally observed in the child's input to appear to the right
of the Head) SVO might be selected with leftward movement of the
verb. (The possibility that Irish selects SVO as a base-generated
order, deriving VSO via V-movement, has been suggested by McKenna
& Wall 1986, as well as Bobaljik and Carnie 1995). These orderings
naturally are derived via the Head-Medial Principle in
the following manner. Again, suppose in the child's input the Complement
tends to be placed to the left of the verb, then, under Kayne's
Head-Medial Principle, the base-generated word ordering requires
the Specifier to be placed to the right of the Head-Verb--yielding
OVS. [19]In sum, we now can maintain the
following two predicted base-generated orders: (i) SVO
and (ii) OVS--with all other orderings being a derived
(from the two) via movement.
(FPH Revisited)
Let us recall a major theoretical premise that has been
discussed in previous chapters, summarized here in the following
manner. Given the absence of parameterization at the early Non-INFL
grammatical stage (assuming FPH to be correct), a number of variables,
governed only by UG, should remain open to the child. In accordance
with the FPH, one might think that such a highly permissive stage
would yield a significant amount of variability. However, such a
claim is too strong and doesn't appropriately compensate for e.g.
frequency of input, individual variation, and so forth. More specifically,
theoretical availability may not have the last say in the matter
and nonlinguistic motivations for building-up language
strategies might over-ride more linguistic motivations--though a
seemingly similar path (on the surface) might be seen as developing
in both events, producing more-or-less the same outcomes. Such an
idea regarding (word order) linearization is summed-up rather nicely
by Martin Atkinson and, I think, is worthy here of full quotation
(below):
[C]onsider the fact that the child is surrounded by utterances
in which the subject precedes the predicate. Of course, for the
adult...the order is linguistically motivated via parameterization
of direction of nominative Case-assignment from I(nfl). By assumption,
the child has no I(nfl) yet, so this linguistic motivation is
not available to him/her. But I can see no reason why such a child
should not use the language he or she hears to effect the linearization
strategy....[W]hen I enters the child's system enabling the appropriate
parameter to be set, a word order which has been used continues
to be used but is now linguistically motivated. A fundamental
change in the internal system of representation has no overt consequences
for this aspect of the child's linguistic behavior (1992:266).
1.3 The Data
(Subjects)
Early word orders in my English data seem to follow two patterns
of acquisition. In the first 12 files (1;10-2;6) there seem to be
even distributions between the rates of SV and VS (SAS) structures.
This symmetry peaks at around file 10 where a 17-to-16 frequency
count (VS-to-SV) is found--where the relative frequency of multi-constituent
DAS to two-constituent SAS-structures is around 3-to-1. At file
13 however, a greater shift towards the usage of multi-constituency
DAS-structures occurs, resulting in the dominant use of the target
SVO word order. At first glance it seems that the correct setting
of SVX ordering at this stage is indeed mostly due to the emergence
of multi-constituent strings--notably, either the Object or the
Adverbial element--whereas a steep decline in non-target VS orderings
results as a consequence. Certain characteristics of VS structures
however pose analytical problems: i.e, one such marked feature of
the VS word order stage is the domineering presence of either the
Copula Verb 'Be', or the Bare Past-Participle (=BPP--Bare in the
sense that no Aux surfaces) in many of the constructions. Examples
of these two marked-VS constructions are given in (7):
(7)
a. all gone Truck (file 2: 1;10)
b. all gone Bottle (file 6: 2;2)
c/d. all done The car/Me (files 7/16)
|
e. is A car (file 5: 2;0)
f. is A duck (file 4: 2;0)
g. (Cat), is A cat (file 3: 1;11) |
It is not entirely clear whether or not these constructions constitute
true VS orderings, as the grammatical-categorical status of the
Noun and participle can be questioned. For instance, in examples
(a-d), the participle might be reinterpreted as having Adjectival
properties, while the Copula in examples (e-g) somewhat complicates
matters--i.e., their argument structures may be analyzed as pseudo
unaccusative verbs (without Subject raising), or that assumptions
might be made regarding a phonologically reduced subject via a consonant
cluster simplification (as in e.g., It is a car, its>is>iz--SVO) [20]etc.
However, alongside such problematic constructions lie a considerable
number of straightforward VS constructions containing main verbs.
The following token examples of VS structures are given in (8) below
with analysis in (8-prime). (Note that these examples span the range
of ages from 2;6-3;2--demonstrating that word order variance seems
to solely rely on the number of arguments which project and no other
criteria such as e.g. the functional IP-stage. Hence, such a diverse
range of SAS VS projections suggests that the SAS vs. DAS distinction,
as laid out here, functions rather in an isolated manner, quite
untypical of 'benchmark' criterion which aid in establishing the
child's overall stage of development.
(NB. This gives the flavor of saying that any talk of "stages"--noting
that by file 14 we have surely entered into the IP-stage--simply
doesn't adequately describe the nature of the word order variance
being produced here, and that we must reconsider the overall notion
of an SAS-phase overlapping onto an otherwise functional IP-stage).
(8)
a. kick baby (file 14: 2;7)
b. run baby (file 14: 2;7)
c. all break My bike (file 22: 3;0)
d. open me (file 19: 2;10)
e. eat baby (file 12: 2;6)
f. cook daddy (file15: 2;8)
g. help me (file 23: 3;2) |
h. kick me (file 15: 2;8)
i. eat me (file 17: 2;8)
j. work bike (file 22: 3;0)
k. broke tree (file 23: 3;0)
l. want me (file 18: 2;9)
m. hurt car (file 16: 2;8)
n. go plane (file17: 2;8) |
Next
Page>>
Chapter
3 Notes
[1] It is now commonly suggested that missing arguments
are present at (LF). Such missing arguments (by extension of Rizzi)
may be labelled as implicit and are considered lexically
saturated and need not project into surface syntax (cf. Rizzi
1986, Radford 1990). L. Bloom notes that Subject+Complement constructs
also appear at this stage. (Such examples would have to be considered
as DAS constructs.)
[2] Following assumptions based on Predication, Tsimpli
rules out the possibility that VSO is a base-generated order.
Hence, Welsh and Irish word orderings (two such languages with
strict VSO) must therefore be derived via Verb raising to a functional
head.
[3] Of course, Kayne's treatment of a universal Spec>Head>Comp
ordering may counter this--see (_3.1.2) below for discussion.
[4] One potential problem with this remark, however, is
that it presupposes an unwanted separation between morphology
(be it in its rich or weak form) and syntax--subsequently raising
questions regarding the nature and learnability of syntax.
[5] The Morphological Uniformity Hypothesis which
likens e.g., Chinese to Italian is a good example in this respect
and demonstrates how problems can arise when defining a morphologically
rich language (cf. Jaeggli, O & Hyams, N. 1988)
[6] For Braine (1976) target word order was seen to be fixed
immediately after the first onset of the verb's second argument
projection. Tsimpli (cf. Bowerman 1990:1275) however reports that
target SVO word order becomes fixed just 6-8 weeks into the first
onset of the verb's second argument.
[7] A linear ordering has three defining properties--only
one of them is important to us here: Linear ordering is total;
that is, it must cover all the members of the set: for all distinct
x,y, either xLy or yLx (Kayne 1994:4). Thus the structure
in (3 prime) is ruled out on the basis that the set [j,m,p] does
not properly restrict the possible free ordering of the two terminals
m and p within the set.
[8] I assume here that no canonical ordering holds with
respect to thematic argument structures (see Pinker 1984 and Bowerman
1990 for further discussion).
[9] As later noted in (_3.1.7), such constructions are in
no way tidy and without problems. The very essence of a binary
merger operation starting from the lexicon [a + b] and projected as a primary internal projection (=X') is put into jeopardy:
Secondly, it still remains rather unclear how two-place
predicates with understood arguments would be maintained within
such a proto-type XP. One possibility might be that all understood
arguments missing from the syntax are lexically saturated and
thus implicit (cf. Rizzi).
[10] Margaret Deuchar (ms. 1993) has recently proposed a
radical alternative for a two word stage showing variable word
orders. She claims that children's earliest X-bar syntax reflect
the asymmetry of daughters, but not the existence of mothers (hence,
only X'). In other words, the child merely utilizes an innately
given Head vs. Non-Head distinction for hierarchical structure.
Deuchar's X-bar syntax, at this two-word stage, would therefore
only mark Head and non-Head presumably within a flat structure:
The major problem with the above reduced flat structure
is twofold: (i) It proposes massive and irrecoverable Discontinuity
with adult grammar. For this reason, the view is commonly held
that XP hierarchy is innately given to the child. (ii) Regarding
my own data, though there indeed exists such a two-word (SAS)
phase (with mixed orders), the phase however, at times, proceeds
alongside a three-word phase (DAS) with fixed SVO order (found
in slightly later files)--demonstrating the total hierarchical
X-bar syntax (i.e., external argument--Spec, Head, and internal
argument--Comp). We therefore conclude
that if XP is fully 'functional'
at this early DAS phase, it must at least be partially 'functional'
at the slightly earlier SAS phase.
[11] There might be some similarities here with earlier suggestions
that Agro/s were strictly mnemonic as placeholders
and contained no substantial differences--i.e., they both shared
collective phi-features (cf. Chomsky 1995:121).
[12] Also see Radford (1990), Atkinson (1992: 181) among
others, for similar views which suggest that lexical items initially
lack categorial labels entirely and thus, for the child, a much
more primitive structure is engaged (e.g.,...[[a]] [[b]]...). Such a structure poses problems for the assumption that
X-bar principles are instantaneous at all levels.
There seems to be some evidence that the mapping of argument
projections do not follow a universal pattern. An example, the
Australian language Warlpiri is given (a language which seems
to map Agent-Theme, Agent-Patient according to Case rather than
to external/internal argument positions within X-bar).
[13] For similar findings see Deuchar (ms.1993). Though we
differ with Deuchar in that we assume (following Stowell 1981:
70) that all Subjects appear as External arguments at the XP
level--establishing the hierarchy between mother (XP) and daughter
(X'). XP here is not fully expressed only in the sense
that the internal argument Complement slot--normally occupied
by the Object--does not project, subsequently forcing the Object
to surface as a superficial subject within Spec.
See
footnote 10.
[14] This might demonstrate that early German grammars--which
indeed show SOV--immediately project at least one Functional
category for the Object to move into (see Clahsen 1994 for discussion)
(Also see note 13.)
[15] Principles of X-bar structure remain consistent and
a mirror image of SVO projects--adhering to the Head Medial
Principle:
[16] In the case of e.g., SOV, the subject might be viewed
here as raising up to some SPEC position to the left of the Head--bypassing
the Complement along its way.
[17] Kayne argues that X-bar syntax should not be considered
as the foremost primitive part of syntactic theory (i.e., of UG).
What is primitive in UG are the properties derived via (LCA):
(1) the need for a phrase to have a Head, (2) the impossibility
for a phrase to have more than one Head, (3) the limitation to
one Specifier per phrase, (4) the limitation to one sister Complement
per Head, (5) the requirement that Complements not be a bare head
(ibid: 131), and (6) the resulting requirement that Heads be placed
Medially (ibid: 35).
[18] This differs from Kayne's stricter account which claims
that since (UG) only allows for SVO ordering, SOV must then be
derived by SVO and the Complement (not the Subject) has raised
to some Specifier position left of the Head (ibid:35).
[19] A specific issue is raised here concerning the status
of German base-generated word order. It would be argued that the
child has considerable SOV input which would render the ordering
similar to Japanese OVS. The main clause V-2 phenomena then would
have to be considered as the driving input for German. In this
sense, German would have to be considered as a SVO base-generated
language.
[20] An implicit expletive subject here might be ruled out
on the basis that the copula would most likely phonetically contract
to it's: /ts/ > /s/ and not /z/.
|